What the Fall 2026 AI Releases Change for PM Firms

Topic: AI Tools | Type: News | By Anush Samiev, COO at Automation Rabbit | Published 2026-09-19

The AI releases that landed this summer were not about smarter answers. They were about access.

Until this year, using AI at a property management firm meant copying a report out of Entrata, pasting it into a chat window, and copying the answer back into a spreadsheet. That loop capped the value at whatever one person could paste in an afternoon. Between July and September, the major vendors shipped products that skip the pasting. The model reads the file, opens the app, and hands back a finished document.

That is a different kind of tool, and it comes with a different set of questions for whoever runs your operation.

What actually shipped

Google has been shipping on the same track with Gemini Enterprise and its Workspace admin controls. The pattern is the same across all three vendors.

The part worth paying attention to

Strip out the model names and one thing is left: these products now reach into the systems where your work already lives.

ChatGPT Work connects to more than 1,400 apps. You describe an outcome, it shows a plan, then it runs for a while and gives you the artifact. Claude went the other direction and put document, slide and design creation inside the chat itself, so a request like "turn these twelve variance reports into an owner summary" produces a document you can send rather than a wall of text you have to reformat.

For a PM firm, that maps onto work your team is doing by hand right now:

• Owner reporting packets assembled from accounting exports
• Weekly leasing summaries built from saved reports
• Lease abstraction into a standard summary format
• Vendor invoice coding and exception flagging
• Board and investor decks that get rebuilt every month

None of that requires a smarter model. It requires a tool that can open the file.

What this does not fix

The gap between a good demo and a working process is almost never the model. It is your data.

Here is a pattern we see on every engagement. A firm exports a leasing report, runs it through an AI tool, and the numbers come back wrong. The model did nothing wrong. The export contained duplicate guest cards, three property naming conventions, and no table anywhere that says how many rentable units each property has. Software of any kind produces a confident wrong answer against that input.

Buildium's 2026 research found AI adoption among property management professionals jumped from 20% to 58% in a year, while only 8% have fully automated workflows. That gap is the data work. Everyone can run a prompt. Almost nobody has cleaned up the source reports enough to trust the output without a human checking it.

Before your firm buys seats, somebody should be able to answer three questions:

1. Which system is the source of truth for unit counts, and who updates it?
2. When the same lead shows up in two reports, which record wins?
3. What does a finished, correct version of this report look like, so we can check the automated one against it?

If those answers do not exist, a new AI subscription buys you faster wrong answers.

The cost question

Pricing has settled into a shape most firms can budget against. Claude Team seats run $25 per user per month billed monthly, or $20 annually, with premium seats at $125 for heavier use, and Enterprise at $20 per seat plus usage. ChatGPT Work is metered by task complexity rather than sold as a separate subscription, and it is included on the higher business tiers rather than the entry ones.

For a 30 person firm, a serious pilot is a few hundred dollars a month. That is not the expensive part. The expensive part is the six weeks your controller spends building something that does not survive her vacation.

Where this gets you in trouble

One caution before anything touches a prospect or resident. AI leasing tools have already produced Fair Housing exposure, including a $2.275 million SafeRent settlement over algorithmic screening outcomes. Under the Fair Housing Act, disparate impact does not require intent. If your chatbot answers one inquiry instantly and routes another to a 48 hour callback, the transcript is the evidence.

Internal work carries none of that risk. Reporting, reconciliation, lease abstraction and vendor workflows are where the safe wins are, and they are also where most of the manual hours sit. Start there, and treat anything resident-facing as a project with legal review attached.

What to do this quarter

Pick one report that gets rebuilt every week by hand. Document exactly how it is built today, including the judgment calls the person makes without thinking about them. Then automate that one thing end to end and run it in parallel with the manual version for a month until the numbers match.

That is a smaller ambition than most AI pitches, and it is the only version that compounds. One report becomes a clean data pipeline, and the second report costs a third as much to build because the pipeline already exists.

We do this as a fractional automation officer engagement for property management firms, which means someone owns the roadmap, says no to the builds that will not pay off, and leaves the client owning the infrastructure. If you would rather run it in house, the sequence above still works. The order matters more than who does it.

Next in this series: a side-by-side look at ChatGPT Work and Claude Cowork for a 30-person property management firm, including where each fits best and what the tradeoffs look like in practice.